MULTI-DOCUMENT SUMMARIZATION USING

SENTENCE CLUSTERING
1
Akash Kamble, 2Vivek Chaudhary, 3Adesh Dhakane, 4Abhijit Ghorpade
1
Student, 2Student, 3Student, 4Student
1
Department of Computer Engineering,
1
Sinhgad Academy of Engineering, Kondhwa(Bk.), Pune, India
1
[email protected],2 [email protected],
3
[email protected], [email protected]

Abstract
Multi-document summarization is used for generating the summary of the documents
which will provide the central idea about the documents in short. As large amount of
information is available on internet, there is a need of system that can provide
informative summary in a short time period. In this paper, we have proposed a system
generating the summary of multiple documents related to a particular topic as well as it
will show summary of particular document. In our proposed system, we use tokenization
and stop-word removal method as a preprocessing step of latent dirichlet allocation
(LDA) model. Our approach also uses sentence clustering and term frequency algorithm
for generating short text summary.

Keywords – Summary, Stop-word removal, Clustering, Term frequency

1. INTRODUCTION

As large number of documents are available on the internet, it seems difficult to get
the required information related to a particular topic. In order to solve this problem, we
have proposed a multiple documents summarization using sentence clustering. Using this
summarization system, user gets the short informative summary related to a particular
topic from multiple documents. Instead of reading number of documents for particular
topic, user can make use of this system to get topic related information. It will save the
reading time of user and user will get the lot of topic related information in short time
period. This system generates the summary of multiple documents of .txt and .pdf file
formats. Also user gets the summary for multiple topics if all that topics present in a
particular document. System also provides the documents related to a searched topic
among the collection of documents before generating the summary. For generating
summary, we use following steps:-

 Firstly we use tokenization and stop-word removal method as a preprocessing

step of LDA algorithm. It is able to remove the useless data such as
prepositions, connectives, etc. to remove the irrelevant information and
extract topic from documents.

 Considering topic related search by the user, we have used term frequency
algorithm to count multiple occurrence of words in a document and then
 accordingly calculate the relevant score of each document for arranging them
as per their relevant score.
 By using sentence level clustering, clusters of sentences are formed from
given documents. It makes the clusters of sentences which are relatively
similar based on a particular word.

2. OBJECTIVES
1. To summarize multiple documents into a short informative summary related to a
particular topic and save the time of users.
2. To generate summary of the multiple documents that can be easily readable and
understandable by the user.
3. To generate the topic related summary of documents supporting more than one file
formats.
4. To provide the documents related to particular topic to the user.
3. RELATED WORK
Wei Li(2010) was proposed a summarization system in which BSU semantic link
network is used to generate the summaries. The approach used in this paper was very
effective for extraction of information and providing good summaries. Baotian Hu(2015)
had written a paper on a text summarization in which summary of Chinese text is
generated using recurrent neural network. Jason Weston(2015) was proposed a model for
the summarization of sentences and provides a short summary.

4. SYSTEM ARCHITECTURE

Figure 4.1
5. SYSTEM ALGORITHM
5.1 Latent Dirichlet Allocation(LDA)
 Latent dirichlet allocation is one of the widely used topic modeling technique.
Topic modeling is used to discover the abstract topics which are in a set of
documents. In data preprocessing steps of LDA, we use tokenization and stop
word removal method for the keyword extraction.
1. Tokenization :- It splits the text data into sentences and generated sentences into
words. Also it converts uppercase letters to lowercase and remove punctuations.
2. Stop-word removal method :- The process of removing stop-words helps to save
the time and reduces the computation. For extraction of keyword, we remove the
stop-words by comparing them with the words in stop-words list given

5.2 .Term Frequency

 Term frequency is used to count multiple occurrences of words which are
occurred in each document.
 Most important words related to document are taken into account and the count
particular term depends on number of presence of words in a specific document.
 Based on this count, documents containing a particular topic are arranged in a
descending order of their relevant score as follows :-
Relevant score:- No of matched words ∕ Total no of words in a document

5.3 Sentence Level Clustering

 In this algorithm, representative documents and sentences including topics are
grouped into multiple clusters using sentence level clustering.
 Sentence clustering avoids the redundancy of sentences. It is domain and
language independent.
 Documents containing a given topic are only considered for the summarization.
Then it makes cluster of sentences based on a given topic and generates the topic
related summary.
 This summary is informative and readable which will save overall time of user

Figure 5.3.1

6. SYSTEM REQUIREMENTS

Hardware requirements:
 Processor : Intel I3 Processor.
 Hard Disk : 40 GB.
 Monitor : 15 VGA Colour.
 RAM : 4 GB
Software requirements:
 Operating system : Windows 7 or Above
 Coding Language : JAVA
 IDE : Eclipse IDE
 Database : SQL
7. RESULT ANALYSIS
In our summarization system, it generates the summary of .txt and .pdf files. So first of
all, we upload .txt and .pdf files and submit to the system. Then the users enter the topic
of their interest and click on “Search” button. After searching for particular topic by the
system, it displays only that documents which include a topic searched by user and if
topic related information does not found in uploaded documents, then system displays
“No Results” to users. Finally when users click on “Summary” button, system generates
informative summary related to the topic given by user.

Figure7.1 Figure 7.2

Figure 7.3 Figure 7.4

8. CONCLUSION AND FUTURE WORK

Multi-document summarization using sentence clustering creates a short summary of
multiple documents. Generated summary is easy to read and understand than existing
systems. It provides important and informative sentences in the summary and removes
repetitive sentences for saving user’s time. Our system generates the summary which
contains the central ideas of given documents and provides the better approach that how
the multiple documents of file formats such as .txt, .pdf can be summarized into short one
related to a particular topic. This summary will provide useful information in a short time
period.
In future work, we plan to make system which can provide the summary of web
pages and multiple file formats such as .exe, .html, .ps, etc. We will integrate and
implement more algorithms related to various summarization approaches with our system
to make our system that can provide the summary of multiple images and videos.

9. REFERENCES
[1] G. Carenini and J. C. K. Cheung, “Extractive vs. nlg-based abstractive summarization
of evaluative text: The effect of corpus controversiality,” in Proceedings of the Fifth
International Natural Language Generation Conference. Association for Computational
Linguistics, 2008, pp. 33– 41.
[2] T. Mikolov, M. Karafi´at, L. Burget, J. Cernock`y, and S. Khudanpur, “Recurrent
neural network based language model.” in Interspeech, vol. 2, 2010, p. 3. DOI:
https://dx.doi.org/10.26808/rs.ca.i8v6.07 International Journal of Computer Application
(2250-1797) Issue 8 Volume 6, Nov.- Dec. 2018 55
[3] K. Filippova, “Multi-sentence compression: finding shortest paths in word graphs,” in
Proceedings of the 23rd International Conference on Computational Linguistics.
Association for Computational Linguistics, 2010, pp. 322–330.
[4] C.-Y. Lin, “Rouge: A package for automatic evaluation of summaries,” in Text
summarization branches out: Proceedings of the ACL-04 workshop. Barcelona, Spain,
2004.
[5] S. Banerjee, P. Mitra, and K. Sugiyama, “Multi-document abstractive summarization
using ilp based multi-sentence compression,” in Proceedings of the 24th International
Conference on Artificial Intelligence. AAAI Press, 2015, pp. 1208–1214.
[6]W. Li, “Abstractive multi-document summarization with semantic information
extraction,” in Proceedings of the 2015 Conference on Empirical Methods in Natural
Language Processing, 2015, pp. 1908–1913.
[7] B. Hu, Q. Chen, and F. Zhu, “Lcsts: A large scale chinese short text summarization
dataset,” arXiv preprint arXiv:1506.05865, 2015.
[8] A. M. Rush, S. Chopra, and J. Weston, “A neural attention model for abstractive
sentence summarization,” arXiv preprint arXiv:1509.00685, 2015.
[9] D. Bahdanau, K. Cho, and Y. Bengio, “Neural machine translation by jointly learning
to align and translate,” arXiv preprint arXiv:1409.0473, 2014.
[10] J. Gu, Z. Lu, H. Li, and V. O. Li, “Incorporating copying mechanism in sequence-to-
sequence learning,” arXiv preprint arXiv:1603.06393, 2016.
[11] S. Bird, “Nltk: the natural language toolkit,” in Proceedings of the COLING/ACL on
Interactive presentation sessions. Association for Computational Linguistics, 2006, pp.
69– 72.
[12] K. Toutanova, D. Klein, C. Manning et al., “Stanford core nlp,” The Stanford
Natural Language Processing Group. Available: http://nlp. stanford.
edu/software/corenlp. shtml. Accessed, 2013.